Main Concept
What is it? What is it about? You can integrate Amazon Bedrock with CloudWatch to monitor and log different metrics and events of the services, this include CloudWatch logs and CloudWatch metrics
Context
Why is it important? How does it relate to other topics? It’s important to monitor resources status and usage to prevent any issue and control the costs.
Key Aspects
CloudWatch logs
- You can log every model invocation and sent to CloudWatch logs or S3 (input and outputs)
- Logs can include text, images.
- You can use CloudWatch Logs Insights to do more analysis and build alerts.
NOTE
S3 is significantly cheaper than CloudWatch Logs for storage—especially for heavy logging.
CloudWatch metrics
- Several metrics are sent from Amazon Bedrock to CloudWatch Metrics, from the model itself and from Guardrails.
- You can set alarms based on thresholds on these metrics.
- For instance, you can use the metric called ContentFilteredCount and set an alarm when the number of matches exceeds a certain number, allowing you to verify that Guardrails are working correctly.
NOTE
We use CloudWatch to monitor infrastructure and operation, however with Amazon SageMaker Model Monitor we can monitor the model quality.
Related Concepts:
Links:
References
Exam Domain
- Domain 5, Task Statement 5.1 - security and privacy considerations for AI systems, best practices for secure data engineering.
- Domain 2, Task Statement 2.3